Synthesize Realistic 3D Medical Images at Scale to Ship Pre‑Trained Models | NVIDIA Technical Blog

This post introduces NV-Generate-MR-Brain, a new model for the synthetic generation of human brain anatomy and structure segmentation built on the MAISI architecture. It extends the MAISI architecture towards scalable, open workflows for synthetic 3D medical imaging generation. NV-Generate-MR-Brain was trained on the multimodal MR-RATE dataset, the world's largest open source multimodal MRI dataset, comprising 100,000 brain MRI studies. The framework includes end-to-end inference code, pretrained weights, and training configurations, released under open source licenses. It addresses limitations of existing medical image synthesis approaches, such as limited generalization, high computational cost, and weak condition alignment, by providing a flexible, foundation model approach adaptable to many downstream tasks and anatomies. NV-Generate-CTMR, a related framework, supports flexible voxel sizes, variable volume dimensions, and whole-body coverage within a single model.